CN114595861A - MSTL (modeling, transformation, simulation and maintenance) and LSTM (least Square TM) model-based medium-and-long-term power load prediction method - Google Patents
MSTL (modeling, transformation, simulation and maintenance) and LSTM (least Square TM) model-based medium-and-long-term power load prediction method Download PDFInfo
- Publication number
- CN114595861A CN114595861A CN202111581634.7A CN202111581634A CN114595861A CN 114595861 A CN114595861 A CN 114595861A CN 202111581634 A CN202111581634 A CN 202111581634A CN 114595861 A CN114595861 A CN 114595861A
- Authority
- CN
- China
- Prior art keywords
- data
- annual
- component
- load
- monthly
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 68
- 238000012423 maintenance Methods 0.000 title description 2
- 238000004088 simulation Methods 0.000 title description 2
- 230000009466 transformation Effects 0.000 title description 2
- 238000007689 inspection Methods 0.000 claims abstract description 46
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 32
- 230000001932 seasonal effect Effects 0.000 claims abstract description 20
- 238000012360 testing method Methods 0.000 claims abstract description 17
- 230000009467 reduction Effects 0.000 claims abstract description 16
- 230000000737 periodic effect Effects 0.000 claims abstract description 14
- 238000012545 processing Methods 0.000 claims abstract description 12
- 238000013528 artificial neural network Methods 0.000 claims abstract description 8
- 230000000306 recurrent effect Effects 0.000 claims abstract description 5
- 238000012216 screening Methods 0.000 claims description 16
- 238000004422 calculation algorithm Methods 0.000 claims description 11
- 238000004519 manufacturing process Methods 0.000 claims description 11
- 230000015654 memory Effects 0.000 claims description 9
- 230000007774 longterm Effects 0.000 claims description 7
- 238000005065 mining Methods 0.000 claims description 6
- 238000005295 random walk Methods 0.000 claims description 6
- 238000004458 analytical method Methods 0.000 claims description 5
- 230000006403 short-term memory Effects 0.000 claims description 5
- 238000004364 calculation method Methods 0.000 claims description 4
- 238000010276 construction Methods 0.000 claims description 4
- 230000008569 process Effects 0.000 claims description 4
- 238000003860 storage Methods 0.000 claims description 4
- 101100356682 Caenorhabditis elegans rho-1 gene Proteins 0.000 claims description 3
- 229910000831 Steel Inorganic materials 0.000 claims description 3
- 230000001413 cellular effect Effects 0.000 claims description 3
- 230000006870 function Effects 0.000 claims description 3
- 239000011159 matrix material Substances 0.000 claims description 3
- 239000002420 orchard Substances 0.000 claims description 3
- 239000010959 steel Substances 0.000 claims description 3
- 238000013277 forecasting method Methods 0.000 claims 3
- 230000010354 integration Effects 0.000 claims 1
- 238000000513 principal component analysis Methods 0.000 description 11
- 238000003973 irrigation Methods 0.000 description 4
- 230000002262 irrigation Effects 0.000 description 4
- 230000008859 change Effects 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 230000007787 long-term memory Effects 0.000 description 3
- OKTJSMMVPCPJKN-UHFFFAOYSA-N Carbon Chemical compound [C] OKTJSMMVPCPJKN-UHFFFAOYSA-N 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 229910052799 carbon Inorganic materials 0.000 description 2
- 238000010219 correlation analysis Methods 0.000 description 2
- 238000007781 pre-processing Methods 0.000 description 2
- 238000003672 processing method Methods 0.000 description 2
- 238000012549 training Methods 0.000 description 2
- 101001095088 Homo sapiens Melanoma antigen preferentially expressed in tumors Proteins 0.000 description 1
- 102100037020 Melanoma antigen preferentially expressed in tumors Human genes 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 238000013213 extrapolation Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 238000009472 formulation Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000006386 neutralization reaction Methods 0.000 description 1
- 238000010248 power generation Methods 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Economics (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- Software Systems (AREA)
- Computational Linguistics (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Marketing (AREA)
- Biophysics (AREA)
- General Business, Economics & Management (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Tourism & Hospitality (AREA)
- Game Theory and Decision Science (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Entrepreneurship & Innovation (AREA)
- Development Economics (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- Primary Health Care (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention relates to a method for predicting power load based on a neural network, which comprises the following steps: collecting annual scale data to be processed and monthly scale data and carrying out correlation test, and determining key influence factors in the annual scale data and the monthly scale data; respectively carrying out seasonal decomposition on the key influence factors, the corresponding annual power maximum load data and the corresponding monthly power maximum load data to obtain corresponding annual data trend components, annual data residual error components, annual data periodic components, corresponding monthly data trend components, monthly data residual error components and monthly data periodic components; then performing co-integration inspection and dimension reduction processing respectively to obtain corresponding components of the LSTM model; inputting each component into an LSTM model to obtain a prediction component of each component; and fitting the prediction component by adopting the self-learning capability of the LSTM recurrent neural network according to the prediction component to obtain a power load prediction value.
Description
Technical Field
The invention relates to the technical field of computers, in particular to a method for predicting an electric power load based on a neural network.
Background
At present, the power load prediction result is related to the formulation of the dispatching operation and the production plan of a power system, and the accurate load prediction result is beneficial to reasonably arranging the production plan of a power generation enterprise and ensuring the safety and the stability of a power grid. With the continuous promotion of the targets of carbon peak-to-peak and carbon neutralization, a large amount of distributed resources are connected into the power system, the load structure of the novel power system is more diversified, the source-load interaction characteristic of the novel power system increases the fluctuation of power requirements, and the difficulty of power grid scheduling is increased. Therefore, a prediction method for more accurately predicting load change and reaction load characteristics is urgently needed to be researched.
The time sequence of the power load and the influence factors thereof often has complex nonlinear characteristics and presents a change rule of a plurality of periods, the load characteristic of a novel power system is more complex, and the access of the flexible load of the electric automobile brings greater load volatility, time variation and randomness by taking the electric automobile as an example. However, the current method for separately predicting the power load data cannot accurately reflect the rules and characteristics of a plurality of change cycles. Therefore, it is necessary to develop a medium-and long-term power load prediction method that can accurately characterize the power load sequence and its influence factors in different periods.
Disclosure of Invention
Aiming at the defects in the prior art, the invention aims to extract a power load sequence and a plurality of periodic variation rules of influencing factors thereof by a multi-period trend decomposition (MSTL) algorithm, and the method can obtain extremely accurate seasonal components with lower calculation cost when extracting a shorter time sequence period, and establishes a power load prediction method based on an MSTL and LSTM combined model by combining the advantages that a long-term and short-term memory (LSTM) recurrent neural network is good at processing various time scale data, thereby improving the accuracy of power load prediction.
In order to achieve the above purposes, the technical scheme adopted by the invention is as follows:
the method for predicting the medium-long term power load based on the MSTL and LSTM models specifically comprises the following steps:
the method comprises the following steps: collecting annual scale data and monthly scale data to be processed;
step two: respectively carrying out correlation test on the year scale data and the month scale data, and determining key influence factors in the year scale data and the month scale data;
step three: performing seasonal decomposition on the key influence factors in the year scale data obtained in the step two and the corresponding annual power maximum load data to obtain corresponding annual data trend components, annual data residual error components and annual data period components;
performing seasonal decomposition on the key influence factors and the corresponding lunar power maximum load data in the lunar scale data obtained in the step two to obtain a corresponding lunar data trend component, a lunar data residual error component and a lunar data period component;
step four: performing co-integration inspection and dimensionality reduction on the corresponding annual data trend component, the annual data residual difference quantity and the annual data period component obtained in the step three to obtain an annual data period input component, an annual data trend input component and an annual data residual error input component of the LSTM model;
performing co-integration inspection and dimension reduction on the lunar data trend component, the lunar data residual component and the lunar data period component obtained in the step three to obtain a lunar data period input component, a lunar data trend and a lunar data residual input component of the LSTM model;
step five: inputting the annual data period input component, the annual data trend input component, the annual data residual error input component, the lunar data period input component, the lunar data trend input component and the lunar data residual error input component into a plurality of LSTM models to obtain an annual data period prediction component, an annual data trend prediction component, an annual data residual error prediction component, a lunar data period prediction component, a lunar data trend prediction component and a lunar data residual error prediction component of the power load prediction value;
step six: and fitting the prediction component obtained in the fifth step by adopting the self-learning capability of the LSTM recurrent neural network to obtain the predicted value of the power load.
On the basis of the scheme, the annual scale data in the step 1 comprise: annual meteorological data, annual economic data, annual power maximum load data;
the monthly scale data includes: monthly power maximum load data and monthly meteorological data;
on the basis of the scheme, the annual economic data comprise: the method comprises the following steps of 1, producing a total value by regions, a first industry, a second industry, a third industry, keeping the total value of the production of the regions constant, keeping the value of the first industry constant, keeping the value of the second industry constant, keeping the value of the third industry constant, power consumption of the whole society, power consumption of residents in daily life, cultivated land area, irrigation of paddy fields, irrigation of irrigated land, dry land, garden land, orchard, urban and industrial and mining land, city, construction town, village, mining land, scenic spot and special land, total value of the production of the regions, first industry, second industry, building industry, third industry, transportation storage and storage of traffic, postal and telecommunication industry, wholesale and retail industry, lodging and catering industry, steel and population proportion of cities and towns;
the annual meteorological data includes: annual average air temperature, annual maximum air temperature, annual minimum air temperature, annual days above 35 ℃ and annual rainfall;
annual power maximum load data includes: 2010 load, 2011 load, 2012 load, 2013 load, 2014 load, 2015 load, 2016 load, 2017 load, 2018 load, 2019 load, 2020 load;
the monthly power maximum load data includes: a load of 1 month, a load of 2 months, a load of 3 months, a load of 4 months, a load of 5 months, a load of 6 months, a load of 7 months, a load of 8 months, a load of 9 months, a load of 10 months, a load of 11 months, a load of 12 months;
the lunar meteorological data includes: monthly average air temperature, monthly maximum air temperature, monthly minimum air temperature, days above 35 ℃ of the month and monthly rainfall;
on the basis of the scheme, the second step specifically comprises the following steps:
calculating a correlation coefficient between annual meteorological data and annual electric power maximum load data and a correlation coefficient between annual economic data and annual electric power maximum load data, and screening out the data with the correlation coefficient larger than 0.8 as a key influence factor and the corresponding annual electric power maximum load data;
calculating a correlation coefficient between the lunar meteorological data and the lunar electric power maximum load data, and screening out the lunar electric power maximum load data and the lunar electric power maximum load data, wherein the correlation coefficient is greater than 0.8 and is used as a key influence factor;
the correlation coefficient calculation formula is shown as formula (1):
wherein Cov (X, Y) is the covariance of the X and Y sequences, σx，σyCorresponding standard deviations for the X, Y sequences.
On the basis of the scheme, the third step specifically comprises the following steps:
performing seasonal decomposition on key influence factors in the annual scale data and corresponding annual power maximum load data by adopting a multi-cycle trend decomposition algorithm to obtain corresponding annual data trend components, annual data residual components and annual data cycle components;
wherein x istThe data at the time t is represented by,
the period component of the annual data is represented,
the trend component of the annual data is represented,
which represents the residual component of the annual data,
comprising a plurality of periodic components
n represents the number of the specific period component obtained by the multi-period trend decomposition algorithm;
performing seasonal decomposition on key influence factors in the monthly scale data and the corresponding monthly power maximum load data by adopting a multi-period trend decomposition algorithm to obtain corresponding monthly data trend components TtResidual component R of lunar datatAnd a monthly data period component StWherein S istComprising a plurality of periodic components St1,St2，...，Stn。
On the basis of the scheme, the processing process of the annual data period component in the four steps comprises the following steps:
performing unit root inspection on the annual data period component by adopting an ADF inspection method, screening the annual data period component which does not meet the stability requirement, and entering the next step of coordination inspection;
the ADF inspection method uses an ADF inspection model as follows:
where α is a constant, δ t is a time trend term, γ ═ ρ -1, Δ ytFor a random walk sequence, Δ yt-iIs Δ ytI-order lag difference term of (u), mutFor random disturbance terms, ytIs a periodic component of the annual data, yt-1Is Δ yt1 order lag difference term of (beta)iIs a linear trend;
the inspection process comprises the following steps: if ρ<1, then ytThe stability is achieved; if ρ is 1, ytA first order single non-stationary sequence; if ρ>1, then ytSequence divergence;
the ADF inspection method starts from the formula (3), and when the formula (3) is not satisfied, the ADF inspection method carries out inspection according to the sequence of the formula (4) and the formula (5); when the original sequence has no unit root, determining the corresponding sequence as a stable sequence, and stopping the test;
performing Jonhenson coordination test on the annual data period component which does not meet the stability requirement to obtain first rule component annual data;
wherein, the expression of the Jonhenson synergy test is as follows:
wherein, Δ ytFor a random walk sequence, ytAs a component of the annual data period, AiIs an endogenous variable coefficient of the VAR model, C is a constant intercept, epsilontIs an error term, pi, gamma is an influence matrix, yt-iIs ytI order lag difference term of (y)t-1Is Δ yt1 order lag difference term of (1);
processing the first rule component year data passing the collaborative inspection by applying a principal cause analysis method, and standardizing the first rule component year data by selecting a Z-score method to obtain a year data period input component Z, as shown in formula (9):
wherein, Z is the normalized sequence, namely the annual data period input component, and K is the annual data of the first rule component.
On the basis of the scheme, the processing method of the annual data trend component, the annual data residual error component, the lunar data trend component, the lunar data residual error component and the lunar data period component is the same as the processing method of the annual data period component.
On the basis of the above scheme, the long-short term memory unit in the LSTM model in step five comprises an input gate itAnd an output gate otForgetting door ftAnd a unit to be learned ptWherein the unit p to be learnedtFor recording all history information up to the present time t and subject to an input gate itOutput gate otAnd forget door ftControl, the input gate itAnd an output gate otAnd forget door ftAre all between 0 and 1.
On the basis of the scheme, the parameters of the long-short term memory unit are updated iteratively through the following formula:
it＝σ(Wxixt+Whiht-1) (10)
ft＝σ(Wxfxt+Whfht-1) (11)
pt＝tanh(Wxpxt+Whpht-1) (12)
ct＝ft×ct-1+it×pt (13)
ot＝σ(Wxoxt+Whoht-1) (14)
ht＝ot×tanh(ct) (15)
in the formula, WxoAnd WhoRepresenting the output gate network parameter to be learned, x1，x2，...，xτTo input the sequence data, xtInput data representing time t, WxiAnd WhiRepresenting the input Gate network parameter to be learned, WxfAnd WhfTo learn network parameters for forgetting gate, WxpAnd WhpFor the cell state to be learned, x represents the point-by-point multiplication operation, ctRepresenting the cellular state of the long-short term memory unit;
when the LSTM model is trained to obtain the stacked long-short term memory network model, a loss function is expressed by the following formula:
wherein m is a statistical parameter, yiIn order to be the true value of the value,
for the predicted value, the statistical parameter m is the mean value of the square sum of the error of the corresponding points of the predicted data and the original data.
The invention has the beneficial effects that:
the invention aims to extract various periodic variation rules of a power load sequence and influence factors thereof through a multi-period-trend-decomposition (MSTL) algorithm, improve the quality of input data, shorten the training time of a model and avoid the problem that the model is possibly over-fitted based on co-integration inspection and principal component analysis. And finally, fitting the multidimensional power load influence factors of different time scales by combining the advantages of the LSTM neural network to obtain a medium-term and long-term power load predicted value with higher accuracy.
Drawings
The invention has the following drawings:
FIG. 1 is a block diagram of the present invention.
FIG. 2 is a flow chart of the main steps of the present invention.
FIG. 3 is a diagram of a long short term memory cell.
FIG. 4 is a graph showing the monthly load decomposition results.
FIG. 5 is a diagram illustrating comparison of predicted values of various methods.
Detailed Description
The technical solutions of the present invention will be described in further detail with reference to the accompanying drawings and examples, but it should be understood that these drawings are designed for illustrative purposes only and thus are not intended to limit the scope of the present invention. Furthermore, unless otherwise indicated, the drawings are intended to be merely conceptual in illustrating the structural configuration described herein and are not necessarily drawn to scale.
The main steps of the invention are shown in figure 2
The method comprises the following steps: collecting year scale data and month scale data to be processed
Collecting annual scale data and monthly scale data to be processed, wherein the annual scale data comprises: annual meteorological data, annual economic data, annual electric power maximum load data;
the monthly scale data includes: monthly power maximum load data and monthly meteorological data;
wherein the monthly power maximum load data comprises: a load of 1 month, a load of 2 months, a load of 3 months, a load of 4 months, a load of 5 months, a load of 6 months, a load of 7 months, a load of 8 months, a load of 9 months, a load of 10 months, a load of 11 months, a load of 12 months;
the lunar meteorological data includes: monthly average air temperature, monthly maximum air temperature, monthly minimum air temperature, days above 35 ℃ of the month and monthly rainfall;
the annual meteorological data includes: annual average air temperature, annual maximum air temperature, annual minimum air temperature, annual days more than 35 degrees and annual rainfall;
the annual economic data include: the method comprises the following steps of 1, producing a total value by regions, a first industry, a second industry, a third industry, keeping the total value of the production of the regions constant, keeping the value of the first industry constant, keeping the value of the second industry constant, keeping the value of the third industry constant, power consumption of the whole society, power consumption of residents in daily life, cultivated land area, irrigation paddy fields, water irrigation fields, dry land, garden land, orchards, town and industrial and mining land, cities, built towns, villages, mining land, scenic spots and special land, total value of the production of the regions, the first industry, the second industry, construction industry, the third industry, transportation storage and postal and telecommunication industry, wholesale and retail industry, lodging and catering industry, steel and urban population proportion;
annual power maximum load data includes: 2010 load, 2011 load, 2012 load, 2013 load, 2014 load, 2015 load, 2016 load, 2017 load, 2018 load, 2019 load, 2020 load;
step two: respectively carrying out correlation test on the annual scale data and the monthly scale data to determine key influence factors in the annual scale data and the monthly scale data
Respectively carrying out correlation test on the annual scale data and the monthly scale data, and screening out key influence factors in the annual scale data, corresponding annual power maximum load data, the monthly scale data and corresponding monthly power maximum load data, wherein the specific steps are as follows:
calculating a correlation coefficient between annual meteorological data and annual electric power maximum load data and a correlation coefficient between annual economic data and annual electric power maximum load data, and screening out a key influence factor with the correlation coefficient being more than 0.8 and corresponding annual electric power maximum load data;
calculating a correlation coefficient between the lunar meteorological data and the lunar electric power maximum load data, and screening out the lunar electric power maximum load data and the lunar electric power maximum load data, wherein the correlation coefficient is greater than 0.8 and is used as a key influence factor;
the correlation coefficient calculation formula is shown as formula (1):
wherein Cov (X, Y) is the covariance of the X and Y sequences, σx，σyCorresponding standard deviations for the X, Y sequences.
Step three: performing seasonal decomposition on key influence factors in the annual scale data obtained in the step two and corresponding annual power maximum load data to obtain corresponding annual data trend components, annual data residual error components and annual data period components;
performing seasonal decomposition on the key influence factors and the corresponding lunar power maximum load data in the lunar scale data obtained in the step two to obtain a corresponding lunar data trend component, a lunar data residual error component and a lunar data period component;
according to annual power maximum load data, performing seasonal decomposition on key influence factors in annual scale data to obtain an annual data trend component, an annual data residual error component and an annual data period component;
the method specifically comprises the following steps:
performing seasonal decomposition on annual data by adopting a multicycle trend decomposition algorithm (MSTL), wherein the annual data comprises annual economic data, annual power maximum load data and annual meteorological data:
the MSTL algorithm is an extension to the period trend algorithm (STL) that can decompose a time series with multiple seasonal periods, as shown in equation (2):
wherein x istThe data at the time t is represented by,
the period component of the annual data is represented,
the trend component of the annual data is represented,
the residual component of the annual data is,
comprising a plurality of periodic components
n denotes the number of the specific periodic component obtained by MSTL decomposition.
Further, a seasonal decomposition is performed on monthly data, the monthly data including: lunar data meteorological data and lunar power maximum load data to obtain lunar data trend component TtResidual component of lunar data RtPeriodic component S of lunar datatWherein S istComprising a plurality of periodic components St1,St2，...， StAnd n, the method is the same as above.
Step four: performing co-integration inspection and dimensionality reduction on the corresponding annual data trend component, the annual data residual difference quantity and the annual data period component obtained in the step three to obtain an annual data period input component, an annual data trend input component and an annual data residual error input component of the LSTM model;
performing co-integration inspection and dimension reduction on the lunar data trend component, the lunar data residual component and the lunar data period component obtained in the step three to obtain a lunar data period input component, a lunar data trend and a lunar data residual input component of the LSTM model;
firstly, performing co-integration inspection and dimension reduction processing on the annual data period component obtained in the step three to obtain an LSTM model annual data period input component; performing co-integration check and dimension reduction processing on the annual data trend component and the annual data residual error component obtained in the third step to obtain an LSTM model annual data trend input component and an annual data residual error input component;
secondly, performing co-integration inspection and dimension reduction processing on the lunar data period component obtained in the third step to obtain an LSTM model lunar data period input component; performing co-integration inspection and dimension reduction processing on the lunar data trend component and the lunar data residual component obtained in the third step to obtain an LSTM model lunar data trend input component and a lunar data residual input component;
the method comprises the following specific steps:
1) for the annual data period component:
screening annual data period components according to an ADF (auto-regressive) inspection method and a Jonhenson collaborative inspection method, and performing dimensionality reduction processing on the annual data period components according to a Principal Component Analysis (PCA) method to obtain an input component of the annual data period of the LSTM model;
performing unit root inspection on the annual data period component by adopting an ADF inspection method, screening the annual data period component which does not meet the stability requirement, and entering the next step of coordination inspection;
the ADF inspection method uses the ADF inspection model as follows:
where α is a constant and δ t is timeTrend term, γ ═ ρ -1, Δ ytFor a random walk sequence, Δ yt-iIs Δ ytI-order lag difference term of (u), mutFor random disturbance terms, ytIs a periodic component of the annual data, yt-1Is Δ yt1 order lag difference term of (beta)iIs a linear trend.
The inspection process comprises the following steps: if ρ<1, then ytStabilizing; if ρ is 1, ytA first order single non-stationary sequence; if ρ>1, then ytThe sequence diverges. Thus, determining whether a sequence is flat can be accomplished by checking that p is strictly less than 1. That is, the initial hypothesis is H0, ρ is 1, and the alternative hypothesis is H1, ρ is<1。
The ADF inspection method starts from the formula (3), and when the formula (3) is not satisfied, the ADF inspection method carries out inspection according to the sequence of the formula (4) and the formula (5); and when the test rejects the null hypothesis, namely the original sequence has no unit root, determining that the corresponding sequence is a stable sequence, and stopping the test.
Performing Jonhenson coordination test on the annual data period component which does not meet the stability requirement to obtain first rule component annual data;
wherein, the expression of the Jonhenson synergy test is as follows:
wherein, Δ ytFor a random walk sequence, ytIs the annual data period component, AiIs an endogenous variable coefficient of the VAR model, C is a constant intercept, epsilontIs an error term, pi, gamma is an influence matrix, yt-iIs ytOf order i hysteresis difference term, yt-1Is Δ yt1 order lag difference term of (1);
in order to avoid the overfitting phenomenon caused by multiple collinearity among component data and excessive characteristic dimensions, the prediction accuracy is influenced by noise and stable prediction cannot be presented, so that dimension reduction and noise filtering are required to be carried out on the component data.
Processing the first rule component year data passing the co-integration test by applying a Principal Component Analysis (PCA), firstly, standardizing the first rule component year data by selecting a Z-score method to obtain a year data period input component Z, as shown in formula (9):
wherein, Z is the normalized sequence, namely the annual data period input component, and K is the annual data of the first rule component.
2) For the annual data trend and residual components:
screening the annual data trend component and the residual error component according to an ADF (auto-regressive model) inspection method, and then performing dimensionality reduction treatment on the annual data trend component and the residual error component according to PCA (principal component analysis) to obtain an LSTM model annual data trend and residual error input component, wherein the method is the same as the method;
3) for the lunar data period component:
screening annual data trend and residual error components according to an ADF (auto-document feature) inspection method, and then carrying out dimensionality reduction on the annual data trend and residual error components according to PCA (principal component analysis) to obtain a plurality of LSTM model monthly data period input components by the same method;
4) for the lunar data trend and residual components:
screening the lunar data trend and residual error components according to an ADF (auto-regressive) inspection method, and then performing dimensionality reduction treatment on the lunar data trend and residual error components according to PCA (principal component analysis) to obtain the lunar data trend and residual error input components of the LSTM model, wherein the method is the same as the above method;
step five: and inputting each input component obtained in the step four into a plurality of LSTM models to obtain each prediction component of the power load prediction value.
Inputting the annual data period input component, the annual data trend input component, the annual data residual error input component, the lunar data period input component, the lunar data trend input component and the lunar data residual error input component into a plurality of LSTM models to obtain an annual data period prediction component, an annual data trend prediction component, an annual data residual error prediction component, a lunar data period prediction component, a lunar data trend prediction component and a lunar data residual error prediction component of the power load prediction value;
the method specifically comprises the following steps:
the specific structure of the LSTM model unit is shown in FIG. 3, and the long and short term memory unit includes an input gate itAnd an output gate otForgetting door ftAnd a unit to be learned ptWherein the unit p to be learnedtFor recording all history information up to the present time t and subject to an input gate itAnd an output gate otAnd forget door ftControl, the input gate itAnd an output gate otAnd forget door ftAre all between 0 and 1.
The parameters of the long and short term memory unit are updated iteratively through the following formula:
it＝σ(Wxixt+Whiht-1) (10)
ft＝σ(Wxfxt+Whfht-1) (11)
pt＝tanh(Wxpxt+Whpht-1) (12)
ct＝ft×ct-1+it×pt (13)
ot＝σ(Wxoxt+Whoht-1) (14)
ht＝ot×tanh(ct) (15)
in the formula, WxoAnd WhoRepresenting the output gate network parameter to be learned, x1，x2，...，xτTo input the sequence data, xtInput data representing time t, WxiAnd WhiPresentation inputDoor to study network parameters, WxfAnd WhfTo learn network parameters for forgetting gate, WxpAnd WhpFor the cell state to be learned, x represents the point-by-point multiplication operation, ctIndicating the cellular state of the long-short term memory unit.
When the LSTM model is trained to obtain the stacked long-short term memory network model, a loss function is expressed by the following formula:
wherein m is a statistical parameter, yiIn order to be the true value of the value,
for the predicted value, the statistical parameter m is the mean value of the square sum of the error of the corresponding points of the predicted data and the original data.
Step six: predicting final load
And aiming at the prediction component obtained in the fifth step, fitting the prediction component by adopting the self-learning capability of the LSTM recurrent neural network to obtain the predicted value of the power load.
Analysis of examples
(1) Basic data
The year scale data and month scale data to be processed are shown in Table 1
TABLE 1 summary of basic data
(2) Correlation test
Taking the maximum load data of electric power from 2010 to 2020 as an analysis target, and screening out the influence factors with strong correlation through data preprocessing and correlation analysis, wherein the influence factors with strong correlation comprise: the total value of regional production (hundred million yuan), the first industry (hundred million yuan), the second industry (hundred million yuan), the third industry (hundred million yuan), the first industry and the like are unchanged in price.
Taking the maximum load data of electric power from 1 month to 12 months in 2010 to 1 month to 12 months in 2020 as an analysis target, and screening out influence factors with strong correlation through data preprocessing and correlation analysis, wherein the influence factors with strong correlation comprise: monthly average air temperature, monthly maximum air temperature, monthly minimum air temperature, monthly days above 35 ℃ and monthly rainfall.
(3) Seasonal decomposition
And carrying out seasonal decomposition on 22 groups of related factor data including monthly power maximum load data and annual power maximum load data to obtain trend components, seasonal components 1 and 2 and residual errors of various influencing factors. Taking monthly maximum load data as an example, an exploded view as shown in fig. 4 can be obtained.
(4) Coordination check
Through the coordination inspection, 26 influencing factors such as first industry (hundred million yuan), regional production total value unchanged price, first industry (ten thousand yuan), wholesale and retail industry, lodging and catering industry (ten thousand yuan), annual maximum load, annual load electricity consumption and the like are reserved.
(5) Calculating the final load and the final electric quantity
The final load refers to the annual power maximum load in 2021-2025 years, and the final power consumption refers to the total power consumption of the whole society in 2021-2025 years.
The method utilizes a traditional single-season decomposition prediction method, a single time series decomposition method and the method to predict the maximum load of 91 months from 2010.1 months to 2017.7, compares the maximum load with a true value, and takes a common Mean Absolute Error (MAPE) as a judgment standard. The results are shown in FIG. 5 and Table 2. It can be seen that compared with the conventional time-series extrapolation model and the conventional seasonal decomposition model, the method provided by the invention has the advantages that the prediction accuracy is greatly improved, and the overfitting phenomenon does not occur. In the aspect of monthly maximum load prediction, the optimal error of other methods is 8.73%, while the method is 2.07%, and the prediction precision is greatly improved.
TABLE 2 comparison of monthly prediction errors for various methods
| Method | Prediction error |
| Single season decomposition prediction | 9.87％ |
| Single time series decomposition | 8.73％ |
| MSTL-LSTM combined model | 2.07％ |
Finally, the method takes the data of 2010 to 2019 as a training sample, and can determine the annual maximum load and the virtual predicted value of the power consumption of Ningxia autonomous regions 2021 to 2025, as shown in table 3 and table 4:
TABLE 3 Ningxia region 2021-2025 year maximum load prediction value
| Year of year | Unit of | Maximum load |
| 2021 year old | Ten thousand kW | 1552 |
| 2022 year old | Ten thousand kW | 1664 |
| 2023 year old | Ten thousand kW | 1721 |
| 2024 year old | Ten thousand kW | 1767 |
| 2025 year old | Ten thousand kW | 1798 |
TABLE 4 prediction of 2021-year-old 2025 power consumption in Ningxia region
| Year of year | Unit of | Annual power consumption |
| 2021 year old | Hundred million kWh | 1345 |
| 2022 year old | Hundred million kWh | 1428 |
| 2023 years old | Hundred million kWh | 1470 |
| 2024 year old | Hundred million kWh | 1491 |
| 2025 years ago | Hundred million kWh | 1501 |
Those not described in detail in this specification are within the skill of the art.
Claims (9)
1. A method for predicting medium and long term power load based on MSTL and LSTM models is characterized by comprising the following steps:
the method comprises the following steps: collecting annual scale data and monthly scale data to be processed;
step two: respectively carrying out correlation test on the year scale data and the month scale data, and determining key influence factors in the year scale data and the month scale data;
step three: performing seasonal decomposition on the key influence factors in the year scale data obtained in the step two and the corresponding annual power maximum load data to obtain corresponding annual data trend components, annual data residual error components and annual data period components;
performing seasonal decomposition on the key influence factors and the corresponding lunar power maximum load data in the lunar scale data obtained in the step two to obtain a corresponding lunar data trend component, a lunar data residual component and a lunar data period component;
step four: performing co-integration inspection and dimension reduction on the corresponding annual data trend component, annual data residual error component and annual data period component obtained in the step three to obtain an annual data period input component, an annual data trend input component and an annual data residual error input component of the LSTM model;
performing co-integration inspection and dimension reduction on the lunar data trend component, the lunar data residual component and the lunar data period component obtained in the step three to obtain a lunar data period input component, a lunar data trend and a lunar data residual input component of the LSTM model;
step five: inputting the annual data period input component, the annual data trend input component, the annual data residual error input component, the monthly data period input component, the monthly data trend input component and the monthly data residual error input component into a plurality of LSTM models to obtain an annual data period prediction component, an annual data trend prediction component, an annual data residual error prediction component, a monthly data period prediction component, a monthly data trend prediction component and a monthly data residual error prediction component of the power load prediction value;
step six: and fitting the prediction component obtained in the fifth step by adopting the self-learning capability of the LSTM recurrent neural network to obtain the predicted value of the power load.
2. The MSTL and LSTM model-based medium-and-long term power load prediction method of claim 1, wherein step 1 the year scale data comprises: annual meteorological data, annual economic data, annual power maximum load data;
the monthly scale data includes: monthly power maximum load data and monthly meteorological data.
3. The MSTL and LSTM model-based medium-and-long term power load prediction method of claim 2, wherein the annual economic data comprises: the method comprises the following steps of 1, regional production total value, a first industry, a second industry, a third industry, regional production total value constant price, first industry constant price, second industry constant price, third industry constant price, power consumption of the whole society, power consumption of resident life, cultivated land area, irrigated paddy field, irrigated land, dry land, garden land, orchard, urban and industrial and mining land, city, construction town, village, mining land, scenic spot and special land, regional production total value, first industry, second industry, construction industry, third industry, transportation storage and transportation and post-and-post-telecommunication industry, wholesale and retail industry, lodging and catering industry, steel and urban population proportion;
the annual meteorological data includes: annual average temperature, annual maximum temperature, annual minimum temperature, annual days above 35 degrees and annual rainfall;
annual power maximum load data includes: 2010 load, 2011 load, 2012 load, 2013 load, 2014 load, 2015 load, 2016 load, 2017 load, 2018 load, 2019 load, 2020 load;
the monthly power maximum load data includes: a load of 1 month, a load of 2 months, a load of 3 months, a load of 4 months, a load of 5 months, a load of 6 months, a load of 7 months, a load of 8 months, a load of 9 months, a load of 10 months, a load of 11 months, a load of 12 months;
monthly meteorological data includes: monthly average air temperature, monthly maximum air temperature, monthly minimum air temperature, days above 35 degrees of the month and monthly rainfall.
4. The MSTL and LSTM model-based medium and long term power load prediction method of claim 2, wherein step two specifically comprises:
calculating a correlation coefficient between annual meteorological data and annual electric power maximum load data and a correlation coefficient between annual economic data and annual electric power maximum load data, and screening out the correlation coefficient more than 0.8 as a key influence factor and the corresponding annual electric power maximum load data;
calculating a correlation coefficient between the lunar meteorological data and the lunar electric power maximum load data, and screening out the lunar electric power maximum load data and the lunar electric power maximum load data with the correlation coefficient larger than 0.8 as key influence factors;
the correlation coefficient calculation formula is shown as formula (1):
wherein Cov (X, Y) is the covariance of the X and Y sequences, σx，σyCorresponding standard deviations for the X, Y sequences.
5. The MSTL and LSTM model-based medium and long term power load forecasting method as recited in claim 4, wherein the third step specifically comprises:
performing seasonal decomposition on key influence factors in the annual scale data and corresponding annual power maximum load data by adopting a multi-cycle trend decomposition algorithm to obtain corresponding annual data trend components, annual data residual components and annual data cycle components;
wherein x istThe data at the time t is represented by,
the period component of the annual data is represented,
the trend component of the annual data is represented,
which represents the residual component of the annual data,
comprising a plurality of periodic components
n represents the number of the specific periodic component obtained by the multi-period trend decomposition algorithm;
performing seasonal decomposition on key influence factors in the monthly scale data and the corresponding monthly power maximum load data by adopting a multi-period trend decomposition algorithm to obtain a corresponding monthly data trend component TtResidual component R of lunar datatAnd a monthly data period component StWherein S istComprising a plurality of periodic components St1,St2，...，Stn。
6. The MSTL and LSTM model based medium and long term power load forecasting method as recited in claim 5, wherein the step four middle year data period component processing procedure comprises:
performing unit root inspection on the annual data period component by adopting an ADF inspection method, screening the annual data period component which does not meet the stability requirement, and entering the next step of collaborative integration inspection;
the ADF inspection method uses an ADF inspection model as follows:
where α is a constant, δ t is a time trend term, γ ═ ρ -1, Δ ytFor a random walk sequence, Δ yt-iIs Δ ytI-order lag difference term of (u)tFor random disturbance terms, ytIs a period component of annual data, yt-1Is Δ yt1 order lag difference term of (beta)iIs a linear trend;
the inspection process comprises the following steps: if ρ<1, then ytStabilizing; if ρ is 1, ytA first order single non-stationary sequence;if ρ>1, then ytSequence divergence;
the ADF inspection method starts from the formula (3), and when the formula (3) is not satisfied, the ADF inspection method carries out inspection according to the sequence of the formula (4) and the formula (5); when the original sequence has no unit root, determining the corresponding sequence as a stable sequence, and stopping the test;
performing Jonhenson coordination test on the annual data period component which does not meet the stability requirement to obtain first rule component annual data;
wherein, the expression of the Jonhenson synergy test is as follows:
wherein, Δ ytFor a random walk sequence, ytAs a component of the annual data period, AiIs an endogenous variable coefficient of the VAR model, C is a constant intercept, epsilontIs an error term, pi, Γ is an influence matrix, yt-iIs ytOf order i hysteresis difference term, yt-1Is Δ yt1 order lag difference term of (1);
processing the first rule component year data passing the collaborative inspection by applying a principal cause analysis method, and standardizing the first rule component year data by selecting a Z-score method to obtain an year data period input component Z, as shown in a formula (9):
wherein, Z is the annual data period input component, and K is the first rule component annual data.
7. The MSTL and LSTM model based mid-and-long term power load prediction method of claim 6 wherein said year data trend component, year data residual component, month data trend component, month data residual component and month data period component are processed in the same way as the year data period component.
8. The MSTL and LSTM model based medium and long term power load forecasting method of claim 7 wherein the LSTM model in step five has a long short term memory unit comprising an input gate itAnd an output gate otForgetting door ftAnd a unit to be learned ptWherein the unit p to be learnedtFor recording all history information up to the present time t and subject to an input gate itAnd an output gate otAnd forget door ftControl, the input gate itAnd an output gate otAnd forget door ftAre all between 0 and 1.
9. The MSTL and LSTM model based medium-long term power load prediction method of claim 8, wherein the parameters of the long-short term memory unit are iteratively updated by:
it＝σ(Wxixt+Whiht-1) (10)
ft＝σ(Wxfxt+Whfht-1) (11)
pt＝tanh(Wxpxt+Whpht-1) (12)
ct＝ft×ct-1+it×pt (13)
ot＝σ(Wxoxt+Whoht-1) (14)
ht＝ot×tanh(ct) (15)
in the formula, WxoAnd WhoRepresenting the output gate network parameter to be learned, x1，x2，...，xτFor inputting sequence data, xtInput data representing time t, WxiAnd WhiRepresenting the input Gate network parameter to be learned, WxfAnd WhfTo learn network parameters for forgetting gate, WxpAnd WhpFor the cell state to be learned, x represents the multiplication operation point by point, ctRepresenting the cellular state of the long-short term memory unit;
the LSTM model, when trained to obtain the stacked long-short term memory network model, represents a loss function by the following formula:
wherein m is a statistical parameter, yiIn order to be the true value of the value,
for the predicted value, the statistical parameter m is the mean value of the square sum of the error of the corresponding points of the predicted data and the original data.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111581634.7A CN114595861A (en) | 2021-12-22 | 2021-12-22 | MSTL (modeling, transformation, simulation and maintenance) and LSTM (least Square TM) model-based medium-and-long-term power load prediction method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111581634.7A CN114595861A (en) | 2021-12-22 | 2021-12-22 | MSTL (modeling, transformation, simulation and maintenance) and LSTM (least Square TM) model-based medium-and-long-term power load prediction method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114595861A true CN114595861A (en) | 2022-06-07 |
Family
ID=81814345
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111581634.7A Pending CN114595861A (en) | 2021-12-22 | 2021-12-22 | MSTL (modeling, transformation, simulation and maintenance) and LSTM (least Square TM) model-based medium-and-long-term power load prediction method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114595861A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115481818A (en) * | 2022-10-12 | 2022-12-16 | 大连理工大学 | Medium-and-long-term runoff forecasting method and system based on time sequence decomposition |
| CN116739118A (en) * | 2023-06-27 | 2023-09-12 | 国网江苏省电力有限公司南通供电分公司 | Power load prediction method for realizing error correction mechanism based on LSTM-XGBoost |
| CN117709530A (en) * | 2023-12-13 | 2024-03-15 | 中国司法大数据研究院有限公司 | Case quantity prediction method for eliminating new year influence of lunar calendar |
-
2021
- 2021-12-22 CN CN202111581634.7A patent/CN114595861A/en active Pending
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115481818A (en) * | 2022-10-12 | 2022-12-16 | 大连理工大学 | Medium-and-long-term runoff forecasting method and system based on time sequence decomposition |
| CN115481818B (en) * | 2022-10-12 | 2023-05-30 | 大连理工大学 | Medium-and-long-term runoff forecasting method and system based on time sequence decomposition |
| CN116739118A (en) * | 2023-06-27 | 2023-09-12 | 国网江苏省电力有限公司南通供电分公司 | Power load prediction method for realizing error correction mechanism based on LSTM-XGBoost |
| CN117709530A (en) * | 2023-12-13 | 2024-03-15 | 中国司法大数据研究院有限公司 | Case quantity prediction method for eliminating new year influence of lunar calendar |
| CN117709530B (en) * | 2023-12-13 | 2024-06-04 | 中国司法大数据研究院有限公司 | Case quantity prediction method for eliminating new year influence of lunar calendar |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113962364B (en) | Multi-factor power load prediction method based on deep learning | |
| Tran et al. | Nature-inspired metaheuristic ensemble model for forecasting energy consumption in residential buildings | |
| Imani | Electrical load-temperature CNN for residential load forecasting | |
| Hong et al. | Probabilistic electric load forecasting: A tutorial review | |
| Wei et al. | Prediction of residential district heating load based on machine learning: A case study | |
| Fan et al. | Development of prediction models for next-day building energy consumption and peak power demand using data mining techniques | |
| CN111260136A (en) | Building short-term load prediction method based on ARIMA-LSTM combined model | |
| CN114595861A (en) | MSTL (modeling, transformation, simulation and maintenance) and LSTM (least Square TM) model-based medium-and-long-term power load prediction method | |
| Xiao et al. | Impacts of data preprocessing and selection on energy consumption prediction model of HVAC systems based on deep learning | |
| Moradzadeh et al. | Heating and cooling loads forecasting for residential buildings based on hybrid machine learning applications: A comprehensive review and comparative analysis | |
| Li et al. | Meta-learning strategy based on user preferences and a machine recommendation system for real-time cooling load and COP forecasting | |
| CN111915092A (en) | Ultra-short-term wind power prediction method based on long-time and short-time memory neural network | |
| CN113554466A (en) | Short-term power consumption prediction model construction method, prediction method and device | |
| CN112149890A (en) | Comprehensive energy load prediction method and system based on user energy label | |
| Ajitha et al. | Design and development of Residential Sector Load Prediction model during COVID-19 Pandemic using LSTM based RNN | |
| CN114862032B (en) | XGBoost-LSTM-based power grid load prediction method and device | |
| Chen et al. | Short-term wind power forecasting using mixed input feature-based cascade-connected artificial neural networks | |
| CN115759415A (en) | Power consumption demand prediction method based on LSTM-SVR | |
| Zhao et al. | Short-term microgrid load probability density forecasting method based on k-means-deep learning quantile regression | |
| CN115545333A (en) | Method for predicting load curve of multi-load daily-type power distribution network | |
| CN107895211A (en) | A kind of long-medium term power load forecasting method and system based on big data | |
| CN114819395A (en) | Industry medium and long term load prediction method based on long and short term memory neural network and support vector regression combination model | |
| Alharbi et al. | Short-term wind speed and temperature forecasting model based on gated recurrent unit neural networks | |
| CN116826745A (en) | Layered and partitioned short-term load prediction method and system in power system background | |
| Ning et al. | Probabilistic short-term power load forecasting based on B-SCN |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |